Awesome

Safe Multi-Agent Isaac Gym Benchmark (Safe MAIG)

Safe Multi-Agent Isaac Gym benchmark is for safe multi-agent reinforcement learning research.

The README is organized as follows:

• About this repository
• Installation
  • Pre-requisites
  • install this repo
• Running the benchmarks
• Select an algorithm
• Select tasks
  • HandOver Environments
    • <span id="obs1">Observation Space</span>
    • <span id="action1">Action Space</span>
    • <span id="r1">Rewards</span>
  • HandCatchUnderarm Environments
    • <span id="obs2">Observation Space</span>
    • <span id="action2">Action Space</span>
    • <span id="r2">Rewards</span>
  • HandCatchOver2Underarm Environments
    • <span id="obs3">Observation Space</span>
    • <span id="action3">Action Space</span>
    • <span id="r3">Rewards</span>
  • TwoObjectCatch Environments
    • <span id="obs4">Observation Space</span>
    • <span id="action4">Action Space</span>
    • <span id="r4">Rewards</span>
  • HandCatchAbreast Environments
    • <span id="obs5">Observation Space</span>
    • <span id="action5">Action Space</span>
    • <span id="r5">Rewards</span>

About this repository

This repository is an extention of DexterousHands which is from PKU MARL research team, Safe MAIG contains complex dexterous hand RL environments for the NVIDIA Isaac Gym high performance environments described in the NeurIPS 2021 Datasets and Benchmarks paper.

:star2: This repository is under actively development. We appreciate any constructive comments and suggestions. If you have any questions, please feel free to email <gshangd[AT]foxmail.com>.

Installation

Details regarding installation of IsaacGym can be found here. We currently support the Preview Release 3 version of IsaacGym.

Pre-requisites

The code has been tested on Ubuntu 18.04 with Python 3.7. The minimum recommended NVIDIA driver version for Linux is 470 (dictated by support of IsaacGym).

It uses Anaconda to create virtual environments. To install Anaconda, follow instructions here.

Ensure that Isaac Gym works on your system by running one of the examples from the python/examples directory, like joint_monkey.py. Follow troubleshooting steps described in the Isaac Gym Preview 2 install instructions if you have any trouble running the samples.

install this repo

Once Isaac Gym is installed and samples work within your current python environment, install this repo:

pip install -e .

Running the benchmarks

To train your first policy, run this line:

python train.py --task=ShadowHandOver --algo=macpo

Or run the lines:

chmod +x ./run_experiments.sh
./run_experiments.sh

If you want to turn off the video, set headless as True, e.g.,

python train.py --task=ShadowHandOver --algo=macpo --headless=True

Select an algorithm

To select an algorithm, pass --algo=ppo/mappo/happo/hatrpo as an argument:

python train.py --task=ShadowHandOver --algo=macpo

At present, we only support these four algorithms.

Select tasks

Source code for tasks can be found in dexteroushandenvs/tasks.

Until now we only suppose the following environments:

HandOver Environments

These environments involve two fixed-position hands. The hand which starts with the object must find a way to hand it over to the second hand. To use the HandOver environment, pass --task=ShadowHandOver

<span id="obs1">Observation Space</span>

<span id="action1">Action Space</span>

The shadow hand has 24 joints, 20 actual drive joints and 4 underdrive joints. So our Action is the joint Angle value of the 20 dimensional actuated joint.

<span id="r1">Rewards</span>

Rewards is the pose distance between object and goal, and the specific formula is as follows:

goal_dist = torch.norm(target_pos - object_pos, p=2, dim=-1)

quat_diff = quat_mul(object_rot, quat_conjugate(target_rot))
rot_dist = 2.0 * torch.asin(torch.clamp(torch.norm(quat_diff[:, 0:3], p=2, dim=-1), max=1.0))

dist_rew = goal_dist

reward = torch.exp(-0.2*(dist_rew * dist_reward_scale + rot_dist))

HandCatchUnderarm Environments

These environments again have two hands, however now they have some additional degrees of freedom that allows them to translate/rotate their centre of masses within some constrained region. To use the HandCatchUnderarm environment, pass --task=ShadowHandCatchUnderarm

<span id="obs2">Observation Space</span>

<span id="action2">Action Space</span>

Similar to the HandOver environments, except now the bases are not fixed and have translational and rotational degrees of freedom that allow them to move within some range.

<span id="r2">Rewards</span>

Rewards is the pose distance between object and goal, and the specific formula is as follows:

goal_dist = torch.norm(target_pos - object_pos, p=2, dim=-1)

quat_diff = quat_mul(object_rot, quat_conjugate(target_rot))
rot_dist = 2.0 * torch.asin(torch.clamp(torch.norm(quat_diff[:, 0:3], p=2, dim=-1), max=1.0))

dist_rew = goal_dist

reward = torch.exp(-0.2*(dist_rew * dist_reward_scale + rot_dist))

HandCatchOver2Underarm Environments

This environment is is made up of half ShadowHandCatchUnderarm and half ShadowHandCatchOverarm, the object needs to be thrown from the vertical hand to the palm-up hand. To use the HandCatchUnderarm environment, pass --task=ShadowHandCatchOver2Underarm

<span id="obs3">Observation Space</span>

<span id="action3">Action Space</span>

Similar to the HandOver environments, except now the bases are not fixed and have translational and rotational degrees of freedom that allow them to move within some range.

<span id="r3">Rewards</span>

Rewards is the pose distance between object and goal, and the specific formula is as follows:

goal_dist = torch.norm(target_pos - object_pos, p=2, dim=-1)
# Orientation alignment for the cube in hand and goal cube
quat_diff = quat_mul(object_rot, quat_conjugate(target_rot)
reward = (0.3 - goal_dist - quat_diff)

TwoObjectCatch Environments

These environments involve coordination between the two hands so as to throw the two objects between hands (i.e. swapping them). This is necessary since each object's goal can only be reached by the other hand. To use the HandCatchUnderarm environment, pass --task=ShadowHandTwoCatchUnderarm

<span id="obs4">Observation Space</span>

<span id="action4">Action Space</span>

Similar to the HandOver environments, except now the bases are not fixed and have translational and rotational degrees of freedom that allow them to move within some range.

<span id="r4">Rewards</span>

Rewards is the pose distance between two object and two goal, this means that both objects have to be thrown in order to be swapped over. The specific formula is as follows:

goal_dist = torch.norm(target_pos - object_pos, p=2, dim=-1)
goal_another_dist = torch.norm(target_another_pos - object_another_pos, p=2, dim=-1)

# Orientation alignment for the cube in hand and goal cube
quat_diff = quat_mul(object_rot, quat_conjugate(target_rot))
rot_dist = 2.0 * torch.asin(torch.clamp(torch.norm(quat_diff[:, 0:3], p=2, dim=-1), max=1.0))

quat_another_diff = quat_mul(object_another_rot, quat_conjugate(target_another_rot))
rot_another_dist = 2.0 * torch.asin(torch.clamp(torch.norm(quat_another_diff[:, 0:3], p=2, dim=-1), max=1.0))

dist_rew = goal_dist

reward = torch.exp(-0.2*(dist_rew * dist_reward_scale + rot_dist)) + torch.exp(-0.2*(goal_another_dist * dist_reward_scale + rot_another_dist))

HandCatchAbreast Environments

This environment is similar to ShadowHandCatchUnderarm, the difference is that the two hands are changed from relative to side-by-side posture.. To use the HandCatchAbreast environment, pass --task=ShadowHandCatchAbreast

<span id="obs5">Observation Space</span>

<span id="action5">Action Space</span>

Similar to the HandOver environments, except now the bases are not fixed and have translational and rotational degrees of freedom that allow them to move within some range.

<span id="r5">Rewards</span>

Rewards is the pose distance between object and goal, and the specific formula is as follows:

goal_dist = torch.norm(target_pos - object_pos, p=2, dim=-1)

quat_diff = quat_mul(object_rot, quat_conjugate(target_rot))
rot_dist = 2.0 * torch.asin(torch.clamp(torch.norm(quat_diff[:, 0:3], p=2, dim=-1), max=1.0))

dist_rew = goal_dist

reward = torch.exp(-0.2*(dist_rew * dist_reward_scale + rot_dist))

Cite

If you find the repository useful, please cite the paper:

@article{gu2023safe,
  title={Safe Multi-Agent Reinforcement Learning for Multi-Robot Control},
  author={Gu, Shangding and Kuba, Jakub Grudzien and Chen, Yuanpei and Du, Yali and Yang, Long and Knoll, Alois and Yang, Yaodong},
  journal={Artificial Intelligence},
  pages={103905},
  year={2023},
  publisher={Elsevier}
}